From Runtime as Product to Runtime as Commodity
Microsoft’s move to make its agent runtime free while charging for the surrounding control plane marks a shift where the execution engine becomes a commodity and value concentrates in governance, orchestration, and data platforms that coordinate enterprise AI agents at scale. At Build, Microsoft shipped Scout, its first always-on work agent, on OpenClaw, an open-source runtime originally created as a weekend project. Scout runs continuously on behalf of users, connects to Microsoft 365 data, and reaches external apps through the Model Context Protocol, but the logic loop that keeps it running is now part of the open-source base. Microsoft is also contributing enterprise policy controls back to OpenClaw, helping standardize the runtime layer. This makes the agent runtime free for enterprises and partners, while the paid differentiation shifts to how agents are managed, monitored, and secured across complex environments.
The Microsoft Control Plane Becomes the Product
With the agent runtime free, the Microsoft control plane is positioned as the real product for enterprise AI infrastructure. The pattern mirrors mobile: Android’s base is open, but device management, app stores, and identity are monetized. In the agent world, the control plane covers identity, permissions, policy, monitoring, and fleet-wide governance for agents that run on OpenClaw or other open-source agent tools. Build signals that most of Microsoft’s engineering effort is going into this layer, not the runtime loop. For a finance team that wants an agent to reconcile invoices, the value is less in how the agent executes steps and more in how securely it accesses ledgers, respects policy, and provides auditable trails across systems. That is where enterprises will pay, and where Microsoft plans to differentiate with deep integration into its broader cloud and productivity stack.
Open-Source Agent Tools and Reduced Licensing Friction
By building Scout on OpenClaw and feeding enterprise policy controls back upstream, Microsoft aligns with an open-source agent tools model where the core runtime is shared, extensible infrastructure. This lowers licensing friction for enterprises that want to standardize on a common agent runtime free of proprietary lock-in, while still buying into Microsoft’s higher-level services. According to coverage of Build, roughly five months after launch, OpenClaw has already become a base that Microsoft, Nvidia, and several agent startups are building on at once. That kind of multi-vendor support makes it easier for enterprises to run heterogeneous agent fleets on a common substrate and then plug them into Microsoft’s control plane for governance. The result is a clearer split: agent runtimes become interchangeable plumbing, while operational tools, policy, and observability evolve into the billable control tier.
Fabric, Databases, and the New Enterprise AI Infrastructure Stack
Under this model, data platforms—not runtimes—become the sticky parts of enterprise AI infrastructure. Microsoft Fabric and Microsoft Databases are framed as the unified foundation where agent data, analytics, and AI engines meet. Fabric provides a single environment so that each new agent draws from shared organizational context instead of relearning business rules and data locations in isolation. Rayfin, an open-source SDK and CLI, lets developers and coding agents describe application backends in code and deploy them directly to Fabric, so application data lands in OneLake and is immediately available to the wider Fabric stack. In parallel, Azure HorizonDB delivers a PostgreSQL‑compatible database tuned for AI-powered applications, with cloud-scale storage and compute for low-latency, transactional workloads. Together, these services form the monetizable layer for data and AI workflows that sits above a free agent runtime but below the control plane and application experiences.
What This Means for Enterprise AI Infrastructure Pricing
For enterprises, Microsoft’s strategy clarifies where they will spend: not on agent runtimes, but on the Microsoft control plane, Fabric, and AI-optimized databases that orchestrate and store agent activity. This resembles the open-source infrastructure pattern where Kubernetes or Linux are free, while observability suites, policy engines, and managed platforms command premium pricing. Organizations building agentic apps can standardize on an agent runtime free of per-instance licensing while focusing budgets on governance, compliance, and operational reliability at the control-plane level. In practice, that means fewer licensing negotiations around the runtime and more deliberate choices around which control and data planes will host long-lived, always-on agents. Microsoft is betting that once data, context, and operational guarantees live inside Fabric and Microsoft Databases, customers will anchor their enterprise AI infrastructure there, even as the underlying runtimes remain open and interchangeable.
